Mechanics
A claw machine in a shopping mall takes about a USDT a play. It runs whether anyone is watching, and in a good month it clears more than the rent on the floor it stands on.
The person who owns four of them cannot borrow against a fifth.
A $50,000 loan costs a bank almost what a $5 million loan costs to assess. Same paperwork, same hours of somebody's time. Only one of them is worth doing.
So the small ones do not get done. The financing gap for small business across emerging markets runs to 5.7 trillion dollars, and the machines at the bottom of that number are often the most reliable earners in the whole market.
The record is worse again. Nobody outside the business has ever audited what these machines take. An owner can tell a lender what they earned, and backing that up costs more than the loan is worth.
Then the machines started reporting for themselves.
The cost that kept them unfundable was the cost of finding out. A machine that publishes what it did is cheap to find out about, continuously, without anyone driving to a mall to count. The credit follows the machine rather than the person who owns it, which is how equipment has always been financed, and the assessment was the only thing standing in the way.
A laundromat runs a 20 to 35% net margin and returns 30 to 40% unleveraged. Those numbers clear any lending rate with room left over. They were never the problem. The problem was that an owner could not prove them to a stranger for less than the loan was worth.
Proving it to a stranger takes five things, and we did not pick partners so much as work out what each one required.
All of it had to happen in one place. Compute got tokenised, and so did power. Every financial asset went early and went in volume: equities, treasuries, funds, a claim on a claim. We went looking for tokenised machines and found none. Steel was the one nobody brought.
200 machines are running now, and the distributions on them will settle in USDT every month once they go live. The structure behind them is older. DualMint has financed 1,310 everyday assets since May 2025 without a default, and that history belongs to laundromats in Vietnam and air conditioning units in Hong Kong. This fleet inherits it.
Nothing in the method is specific to arcades. A machine qualifies when it earns and reports its performance, and the machines being built now report far more than a claw machine ever will.
We have to earn the receipts in public, on a schedule, and publish the dates before the results.
Which returns us to the mall.
The machine does not know any of this happened. It takes the next USDT and waits for the next player to win or lose. The only thing that changed is that it can tell somebody now, and somebody can check.
Comment on our pinned X post to confirm you read the mechanics →